Forecast accuracy: forecast vs observations
ICON forecast (nearest point, lapse-rate corrected) vs METAR/SYNOP at valid time. MAE — mean absolute error; bias — mean (forecast − observed); ±2° — share within 2 °C.
· 2992137 pairs, 7957 stations, since 24.08 17:00 UTC
Model accuracy — temperature (short range)
| Model | MAE T, °C | Bias T | N |
|---|
| DWD NWP 🥇 | 1.33 | -0.12 | 43963 |
| ECMWF | 1.6 | -0.46 | 47724 |
| GFS | 1.92 | -0.25 | 47724 |
Lower MAE = more accurate. Matched METAR/SYNOP, short range; at 3-7 days the ranking usually differs (ECMWF stronger). Updated hourly.
All stations
| Lead time | N | MAE T, °C | Bias T | ±2° | MAE wind, m/s | MAE P, mmHg |
|---|
| 1–6 h | 109555 | 1.3 | -0.2 | 81% | 1.2 | 0.6 |
| 7–12 h | 115113 | 1.3 | -0.2 | 79% | 1.3 | 0.6 |
| 13–24 h | 230205 | 1.3 | -0.2 | 79% | 1.3 | 0.7 |
| 25–48 h | 459933 | 1.4 | -0.2 | 77% | 1.3 | 0.7 |
| 49–72 h | 395089 | 1.5 | -0.2 | 74% | 1.3 | 0.8 |
| 73–120 h | 35681 | 1.7 | -0.4 | 69% | 1.4 | 0.9 |
SYNOP stations
| Lead time | N | MAE T, °C | Bias T | ±2° | MAE wind, m/s | MAE P, mmHg |
|---|
| 1–6 h | 55702 | 1.3 | -0.3 | 81% | 1.2 | 0.7 |
| 7–12 h | 55775 | 1.3 | -0.3 | 79% | 1.3 | 0.7 |
| 13–24 h | 111529 | 1.4 | -0.3 | 79% | 1.3 | 0.7 |
| 25–48 h | 222581 | 1.4 | -0.3 | 76% | 1.3 | 0.8 |
| 49–72 h | 185959 | 1.5 | -0.3 | 74% | 1.3 | 0.8 |
| 73–120 h | 14263 | 1.7 | -0.5 | 68% | 1.4 | 0.9 |
Airports (METAR)
| Lead time | N | MAE T, °C | Bias T | ±2° | MAE wind, m/s | MAE P, mmHg |
|---|
| 1–6 h | 53853 | 1.2 | -0.2 | 81% | 1.3 | 0.5 |
| 7–12 h | 59338 | 1.3 | -0.2 | 79% | 1.3 | 0.6 |
| 13–24 h | 118676 | 1.3 | -0.1 | 79% | 1.3 | 0.6 |
| 25–48 h | 237352 | 1.4 | -0.1 | 77% | 1.3 | 0.7 |
| 49–72 h | 209130 | 1.5 | -0.1 | 74% | 1.3 | 0.7 |
| 73–120 h | 21418 | 1.7 | -0.2 | 69% | 1.4 | 0.8 |
By country (≤24 h)
| Country | N | MAE T | Bias |
|---|
| Andorra | 68 | 5.9 | -5.9 |
| Liechtenstein | 160 | 3.9 | -3.9 |
| Tajikistan | 760 | 2.9 | -2.3 |
| Nepal | 258 | 2.6 | -2.6 |
| Svalbard and Jan Mayen | 166 | 2.5 | -2.3 |
| Sao Tome and Principe | 86 | 2.4 | -2.4 |
| Reunion | 252 | 2.4 | -1.8 |
| Niger | 790 | 2.4 | +2.1 |
| Belize | 102 | 2.4 | +0.3 |
| Haiti | 73 | 2.3 | -2.3 |
| Bosnia and Herzegovina | 998 | 2.3 | -1.5 |
| Honduras | 799 | 2.2 | -1.6 |
| Aland Islands | 326 | 2.2 | +0.0 |
| Namibia | 849 | 2.1 | -1.4 |
| Burkina Faso | 720 | 2.0 | +1.3 |
| Switzerland | 5574 | 2.0 | -1.2 |
| Saint Kitts and Nevis | 98 | 2.0 | -1.8 |
| Ecuador | 1038 | 2.0 | -1.3 |
| Greenland | 166 | 2.0 | -1.9 |
| Oman | 984 | 1.9 | -1.0 |
Largest errors (≤24 h)
| Station | Country | N | MAE T | Bias |
|---|
| Svobodnyj S31445 | Russia | 8 | 14.4 | +14.4 |
| Madri-Colmenar S08219 | Spain | 68 | 13.8 | +13.8 |
| Darvaz S38856 | Tajikistan | 28 | 13.2 | -13.2 |
| Ishkashim S38957 | Tajikistan | 28 | 11.3 | -11.3 |
| Turgen S36889 | Kazakhstan | 28 | 10.9 | +10.9 |
| Bunji S41518 | Pakistan | 12 | 10.3 | -10.3 |
| Vorogovo S23973 | Russia | 24 | 9.9 | +9.5 |
| Rushan S38951 | Tajikistan | 24 | 9.3 | -9.3 |
| Nyingchi S56312 | China | 24 | 9.3 | -9.3 |
| Khorog S38954 | Tajikistan | 28 | 9.2 | -9.2 |
| Gonbad Ghabus S40735 | Iran | 12 | 9.1 | +9.1 |
| Izana S60010 | Spain | 80 | 9.1 | -9.1 |
| Quetzaltenango S78629 | Guatemala | 12 | 8.8 | +8.8 |
| Houei-Sai * S48926 | Laos | 16 | 8.7 | +8.7 |
| Kangding S56374 | China | 20 | 8.2 | -8.2 |
Typical causes: mountain/coastal siting, elevation mismatch, nocturnal inversions, urban heat island.
